Scholarlink

<p>The growing complexity of educational pathways and the increasing diversity of learner profiles have strengthened the need for recommender systems that are not only accurate but also adaptive and explainable. Neuro-symbolic approaches respond to this need by combining explicit symbolic reasoning (e.g., rules, ontologies, and logic) with neural models particularly graph neural networks (GNNs) operating over knowledge graphs (KGs). This systematic literature review analyzes how neuro-symbolic educational recommender systems represent, operationalize, and integrate symbolic and neural components across educational recommendation tasks, and it identifies the methodological and conceptual challenges that remain. Following PRISMA 2020 guidelines, we reviewed 21 peer-reviewed studies published from 2019 to January 2026, retrieved from major scientific databases. The included works were analyzed across five dimensions: recommendation task, target users, knowledge representation, symbolic reasoning mechanism, and fusion strategy. The results show a clear shift toward hybrid KG-centered architectures, while revealing persistent limitations, including shallow or weakly operationalized symbolic reasoning, limited explainability evaluation, static and domain-bounded knowledge graphs, and insufficient integration of pedagogical or policy-driven constraints. Overall, this review provides a structured synthesis of current practices, highlights key research gaps, and outlines actionable directions to support next-generation educational recommender systems that better balance adaptability, semantic grounding, and transparency.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Toward transparent and adaptive educational recommender systems: a systematic review of neuro-symbolic knowledge graph approaches

  • 期刊论文

  • 发表日期：

    2026年7月23日

查看全文
  • Zineb Elkaimbillah,
  • Mohamed Khoual,
  • Zineb Mcharfi,
  • Mahmoud El Hamlaoui

摘要

The growing complexity of educational pathways and the increasing diversity of learner profiles have strengthened the need for recommender systems that are not only accurate but also adaptive and explainable. Neuro-symbolic approaches respond to this need by combining explicit symbolic reasoning (e.g., rules, ontologies, and logic) with neural models particularly graph neural networks (GNNs) operating over knowledge graphs (KGs). This systematic literature review analyzes how neuro-symbolic educational recommender systems represent, operationalize, and integrate symbolic and neural components across educational recommendation tasks, and it identifies the methodological and conceptual challenges that remain. Following PRISMA 2020 guidelines, we reviewed 21 peer-reviewed studies published from 2019 to January 2026, retrieved from major scientific databases. The included works were analyzed across five dimensions: recommendation task, target users, knowledge representation, symbolic reasoning mechanism, and fusion strategy. The results show a clear shift toward hybrid KG-centered architectures, while revealing persistent limitations, including shallow or weakly operationalized symbolic reasoning, limited explainability evaluation, static and domain-bounded knowledge graphs, and insufficient integration of pedagogical or policy-driven constraints. Overall, this review provides a structured synthesis of current practices, highlights key research gaps, and outlines actionable directions to support next-generation educational recommender systems that better balance adaptability, semantic grounding, and transparency.